In the world of robotics there are multitude of algorithms that enable a robot to operate; however, there is need for an optimal solution for enabling said robot to detect an object in its surroundings. We approached this problem using a network designed using CNNs and Conv-LSTM. This helps the network refer to previous images as a sequence to classify what motion the observed object is performing in a 3-dimensional space in conjunction to the objects state in that space. The robot is able to run the network using a small amount of computational power without sacrificing accuracy and speed. This network has the ability to adapt to new and different environments.
